Redirecting STDOUT in child process

Question

Have a Parent process which spawns multipe child process via fork. I want the log files by the parent and child process to be separate. The Problem is child process STDOUT gets redirected into the parent log file as well as the child log file. Not sure what i need to change to avoid child process log message to get into the parent log file. Also i dont understand in the below setEnvironment function the purpose of creating OUT and ERR file handle. This is a existing code so i kept as it is. In the parent process and child process i set the variable $g_LOGFILE to contain different file name so that separate log files are created. Also i call setEnvironment function in both parent and child process. I tried by closing STDOUT,STDERR,STDIN in the child process and calling the setenvironment but it was not working properly.

sub setEnvironment()
{   

  unless ( open(OUT, ">&STDOUT") )
   {
          print "Cannot redirect STDOUT";
          return 2;
    }
    unless ( open(ERR, ">&STDERR") )
    {
          print "Cannot redirect STDERR";
          return 2;
    }


  unless ( open(STDOUT, "|tee -ai $g_LOGPATH/$g_LOGFILE") )
  {
          print "Cannot open log file $g_LOGPATH/$g_LOGFILE");
          return 2;
   }
   unless ( open(STDERR, ">&STDOUT") )
   {
                print  "Cannot redirect STDERR");
                return 2 ;
    }
    STDOUT->autoflush(1);

} 


####################### Main Program ######################################

    $g_LOGFILE="parent.log";

  while ($file = readdir(DIR))
 {  
     my $pid = fork;
     if ( $pid ) {

        setEnvironment();
        #parent process code goes here
        printf "%s\n", "parent";
        next;
     }
     $g_LOGFILE="child.log";
     setEnvironment();
     #child code goes here
     printf "%s\n", "child";
     exit;
 }

wait for @pids

Answer 1

All the other answers are correct (PSIalt's in particular) - I'm merely hoping that I can answer with corrected code that is identifiably close to that in the question. The key things to notice:

"|tee -ai..."

The tee commands prints its standard in to its standard out while also printing to the given file. As PSIalt says, removing it is the easiest way to ensure each process' output goes only to the correct file.

setEnvironment() inside loop for parent

The original code is constantly redirecting STDOUT back to the teeed file. Therefore recapturing STDOUT. Given my code below, if you moved setEnvironment to above #parent process code goes here you would see all but one 'Real STDOUT' and 'Real STDERR' actually appearing in parent.log.

Options

The ideal is to remove any reliance on redirecting STDOUT / STDERR for logging. I would have a dedicated log($level, $msg) function and start to move all code to using it. Initially it's OK if it is simply a façade for the existing behaviour - you can simply switch it out when you reach an appropriate threshold of code covered.

If it's a basic script and doesn't produce stupidly large logs, why not just print everything to STDOUT with some prefix you can grep for (e.g. 'PARENT:' / 'CHILD:')?

It's a bit outside the scope of the question, but consider using a more structured approach to logging. I would consider using a CPAN logging module, e.g. Log::Log4perl. This way, the parent and children can simply request the correct log category, rather than mess around with file handles. Additional advantages:

• Standardise output
• Allow reconfiguration on the fly - change logging level from ERROR to DEBUG on a running-but-misbehaving system
• Easily redirect output - no change is needed to your code to rearrange log files, rotate files, redirect to a socket / database instead, etc...

use strict;
use warnings;

our $g_LOGPATH = '.';
our $g_LOGFILE = "parent.log";

our @pids;

setEnvironment();

for ( 1 .. 5 ) {
    my $pid = fork;
    if ($pid) {
        #parent process code goes here
        printf "%s\n", "parent";

        print OUT "Real STDOUT\n";
        print ERR "Real STDERR\n";

        push @pids, $pid;
        next;
    }
    $g_LOGFILE = "child.log";
    setEnvironment();

    #child code goes here
    printf "%s\n", "child";
    exit;
}

wait for @pids;

sub setEnvironment {
    unless ( open( OUT, ">&STDOUT" ) ) {
        print "Cannot redirect STDOUT";
        return 2;
    }

    unless ( open( ERR, ">&STDERR" ) ) {
        print "Cannot redirect STDERR";
        return 2;
    }

    unless ( open( STDOUT, '>>', "$g_LOGPATH/$g_LOGFILE" ) ) {
        print "Cannot open log file $g_LOGPATH/$g_LOGFILE";
        return 2;
    }

    unless ( open( STDERR, ">&STDOUT" ) ) {
        print "Cannot redirect STDERR";
        return 2;
    }
    STDOUT->autoflush(1);
}

child.log:

child
child
child
child
child

parent.log:

parent
parent
parent
parent
parent

STDOUT taken from terminal:

Real STDOUT (x5 lines)

STDERR taken from terminal:

Real STDERR (x5 lines)

Answer 2

It would appear that the intent of the original code is as follows:

1. when the script is started from, say, a terminal, then provide aggregate parent and child output to the terminal
2. additionally, provide a copy of the parent output in parent.log, and a copy of child output in child.log

Note that @Unk's answer is correct as far as 2. goes, and has less moving parts than any code using tee, but fails to achieve 1.

If it is important to achieve both 1. and 2. above, then take your original code and simply add the following at the top of your setEnvironment method:

sub setEnvironment()
{
    if ( fileno OUT )
    {
        unless ( open(STDOUT, ">&OUT") )
        {
            print "Cannot restore STDOUT";
            return 2;
        }
        unless ( open(STDERR, ">&ERR") )
        {
            print "Cannot restore STDERR";
            return 2;
        }
    }
    else
    {
        unless ( open(OUT, ">&STDOUT") )
        {
            print "Cannot redirect STDOUT";
            return 2;
        }
        unless ( open(ERR, ">&STDERR") )
        {
            print "Cannot redirect STDERR";
            return 2;
        }
    }
    unless ( open(STDOUT, "|tee -ai $g_LOGPATH/$g_LOGFILE") )
    ...

Incidentally, do not forget to also add $pid to @pids if your actual code does not do that already:

  ...
  my $pid = fork;
  if ( $pid ) {
      push @pids, $pid;
      ...

Why and how does this work? We simply want to temporarily restore the original STDOUT immediately before rewiring it into tee, so that tee inherits it as its standard output and actually writes directly to the original STDOUT (e.g. your terminal) instead of writing (in the case of the forked children) through the parent's tee (which is where the child's STDOUT normally pointed to before this change, by virtue of inheriting from the paremnt process, and which is what injected those child lines into parent.log.)

So in answer to one of your questions, whoever wrote the code to set OUT and ERR must have had exactly the above in mind. (I cannot help but wonder whether or not the difference in indentation in your original code is indicative of someone having removed, in the past, code similar to the one you have to add back now.)

Here's what you now get at the end of the day:

$ rm -f parent.log child.log
$ perl test.pl
child
parent
child
parent
parent
child
parent
child
parent
$ cat parent.log
parent
parent
parent
parent
parent
$ cat child.log
child
child
child
child
child

Answer 3

#!/usr/bin/perl 

use strict;
use warnings;
use utf8;
use Capture::Tiny qw/capture_stdout/;

my $child_log = 'clild.log';
my $parent_log = 'parent.log';

my  $stdout = capture_stdout {
        if(fork()){
            my  $stdout = capture_stdout {
                print "clild\n";
            };
            open my $fh, '>', $child_log;
            print $fh $stdout;
            close $fh;
            exit;
        }
        print "parent\n";
   };
            open my $fh, '>', $parent_log;
            print $fh $stdout;
            close $fh;

Answer 4

Ok i tested this code alitle. Here is my sample code. In my code there is similar(not exact) problem: all messages are double-written to childs log file.

So my answers to your questions:

The Problem is child process STDOUT gets redirected into the parent log file as well as the child log file.

This because when you open file with pipe (open(STDOUT, "|tee ...) as a underlying result your process fork() to create child process and then exec into program what you run (tee). Forking(for tee) takes STDOUT of master process so tee will write into parent's logfile. So i think you must revoke using STDOUT handle for master process. Or, second way - remove use of tee - its simplest way.

Also i dont understand in the below setEnvironment function the purpose of creating OUT and ERR file handle.

Seems this is someone's woraround about problem above. You can grep -rE ' \bERR\b' . to search in code if it used or not. Probably someone wanted to save real STDOUT and STDERR to further use.

Answer 5

You can always redirect STDOUT to log file by closing it first and then reopening:

close STDOUT;
open STDOUT, ">", $logfile;

Small downside to this is that once STDOUT is redirected, you will not see any output on terminal during script execution.

If you want parent and child process have different log files, just perform this redirection in both to different log files after fork(), something like this:

print "Starting, about to fork...\n";
if (fork()) {
    print "In master process\n";
    close STDOUT;
    open STDOUT, ">", "master.log";
    print "Master to log\n";
} else {
    print "In slave process\n";
    close STDOUT;
    open STDOUT, ">", "slave.log";
    print "Slave to log\n";
}

I have tested that this works as expected on Linux and Windows.